Competition and regulation in banking

Transcription

1 Competition and regulation in banking Elena Carletti Center for Financial Studies at the University of Frankfurt Merton Str Frankfurt Germany August 11, 2007 I would like to thank Franklin Allen, Hans Degryse, Martin Hellwig, Steven Ongena and Xavier Vives (the editor) for valuable comments and discussions. 1

2 1 Introduction It is well known that banks are special in that they are vulnerable to instability. As the numerous episodes of crises show, banks are fragile and are prone to take excessive risks. Their function as intermediaries between firms and borrowers and the maturity transformation they operate in their asset-liability management make banks play an important role as providers of liquidity to depositors, but also expose them to runs and systemic crises. The greatrelianceondepositsassourceoffundscreatesasevereagencyproblembetweenbanks and depositors in that, being subject to limited liability, banks do not bear the downside risk and have strong incentives to choose risks that are excessive from the viewpoint of depositors. The need of a stable banking sector, together with that of protecting consumers, provides the motivation for the introduction of deposit insurance schemes and lender of last resort facilities. These safety arrangements are effective in pursuing a stable system, but introduce several distortions and call for further regulatory measures, such as capital requirements. Whereas the speciality of the banking system and the need of regulation have attracted much attention both in academic and policy debates, the issue of how competition affects the stability of the system and the effectiveness of regulation is not well understood yet. The desirability of competition in the banking sector has been questioned for a long time. Following the crises of the 1930s, competition was kept limited in an attempt to preserve stability. The process of deregulation in the last decades lifted many of the restrictions on competition, and opened up the possibility for banks to expand their investments in riskier activities and new locations. A new wave of failures followed in the 1980s and The increase in competition following the deregulation wave was regarded as the main reason behind this new instability. As found by Keeley (1990), the decline of banks margins and charter values magnified the agency problem between banks and depositors (or deposit insurance fund), thus inducing banks to take excessive risks and increasing dramatically their failure probabilities. The idea of a negative relationship between competition and stability has been pervasive in the literature since the 1990s, but more recent contributions indicate that the relationship 2

3 is much more complex. What are the trade-offs between competition and stability? How does competition affect the vulnerability of banks to runs and systemic crises, and their incentives to take risk? How does competition influence the effectiveness of the regulatory tools aiming at preserving stability? Can regulation correct the potential negative effects of competition on stability? This paper aims at providing insights to these questions by reviewing the literature on competition, stability and regulation in banking. We start with looking at these issues separately. First, we briefly describe the reasons behind the risk of instability in the banking sector, and the need of regulation. Following what already mentioned above, we distinguish between sources of instability on the liability side (runs and systemic crises) and on the asset side (excessive risk taking), and discuss how regulation can help achieving a stable system. Then, we analyze how competition operates in this sector. The main conclusion is that, as often argued, the standard competitive paradigm is not appropriate for the banking industry. The presence of important markets failures changes dramatically the nature of competition and its outcome. Asymmetric information, switching costs and network externalities create entry barriers and allow banks to retain some market power in the form of informational rents or enhanced differentiation. Interestingly, this literature proceeds by taking the behavior of agents as exogenous. There is no concern for banks incentives to take risk or depositors desire to run prematurely. The only focus in on how the competitive mechanism operates in the presence of market failures. This is averydifferent approach from the stability literature, which instead focuses exclusively on the behavior of agents within the context of agency theory. To understand better the link between competition and stability, we then review the literature addressing how competition affects the fragility and the risk taking problem, as well as the need of regulating the sector. Surprisingly, the issues have not been studied as extensively as one might expect. Despite a growing interest, the literature is still rather limited and inconclusive on many aspects of the trade-off between competition and stability. What emerges is that, whereas the literature on stability is centered around banks vulnerability to bank runs and systemic crises, most of the contributions analyzing the impact of competition 3

4 on stability have instead addressed the impact of competition on banks incentives to take risk, and the possibility of correcting its perverse impact through appropriate regulatory measures. Besides the limited focus, the literature is also still inconclusive. Whereas the prevailing view is that competition worsens the risk taking problem because lower margins and charter values increase the attractiveness of risky investments, some recent contributions have shown that competition may actually lead to the opposite result of improving the risk of banks portfolios once specific features of the banking system, such as the relationship with borrowers or banks monitoring function, are explicitly taken into account. Regulation may help mitigating the trade-off between competition and stability, as long as such trade-off exists. But how to design regulation appropriately? Again, the literature is inconclusive. Despite there seems to be consensus on the negative effect of flat deposit insurance premia, the results are split for what concerns the effectiveness of capital regulation. While this seems effective in some contexts, it needs to be complemented by direct restrictions of competition such as interest rate ceilings in others. Overall, what emerges is again the need of further attention and research on the impact of competition on stability as well as on the appropriate design of regulation. The rest of the paper proceeds as follows. Section 2 introduces the issue of bank stability, distinguishing between vulnerability to runs and systemic crises, and excessive risk taking. Section 3 reviews the contributions on the functioning of competition in the presence of asymmetric information, switching costs and network externalities. Section 4 analyzes more deeply the link between competition and stability, and in particular the link between market structure and financial fragility, and between market structure and excessive risk taking. Section 5 looks at the impact of regulatory tools on the trade-off between competition and stability. Section 6 concludes. 2 Bank instability and the need of regulation It is well known that banks are special because they are more vulnerable to instability than firms in other sectors, and also because less wealthy people may hold some non-negligible 4

5 share of their wealth in various forms of bank deposits. The potential instability of the banking system and the need of consumer protection are the fundamental rationales behind the introduction and development of regulation. The course of events and in particular the US experience suggest two possible connotations of the term instability : the crises occurred in the 1930s show that the banking system is fragile since it is vulnerable to runs and panics; the massive distress which came to light in the 1980s and 1990s demonstrates that intermediaries may have strong incentives to assume excessive risk and that, as a result, the system has a high probability of failure. Bank fragility: individual runs and systemic crises 1 Intermediaries emerge as a response to the imperfection and incompleteness of financial markets. In an economy characterized by asymmetric information and uncertainty, intermediaries are valuable because they have economies of scale in producing information and provide insurance to depositors who are uncertain in their timing of consumption. Information production and insurance provision are the two main characteristics of bank specificity, but are also the sources of their fragility. The informational asymmetries existing between banks, borrowers and depositors, and the maturity transformation that banks operate by investing short-term deposits in long-term assets expose banks to the possibility of runs. Banks offer depositors demandable contracts which allow depositors to withdraw a fixed amount on demand. If the total value of the early withdrawals exceeds the amount available from short-term investments, a run originates and the bank has to sell its illiquid assets. This illiquidity problem may turn into insolvency and force the premature liquidation of the bank if no assets are left after satisfying the early withdrawals. To illustrate the basic mechanism triggering a run, consider a three-date economy, with one bank operating under perfect competition and raising funds from a continuum of depositors of measure one. The bank invests a fraction M in a short-term asset and a fraction 1 M in a long-term asset. The former simply transfers the unit invested from date 0 to 1 The literature on invidual runs and systemic crises is vast. We describe here only a few contributions. Excellent broader reviews are contained in Gorton and Winton (2003) and De Bandt and Hartmann (2002). 5

6 date 1, while the latter yields R>1 at date 2 and <Rif interrupted prematurely at date 1. Depositors are all ex ante identical, but face a preference shock at date 1. A fraction t of thembecomesoftype1(earlytype)andwishestoconsumeatdate1,whiletheremaining fraction 1 t turnstobeoftype2(latetype)andprefers consuming at date 2. Depending on the specific assumptions on the return of the long-term investment and on the structure of the preference shocks, the rationale behind depositors withdrawal differs and runs can be either irrational or information-induced events. Following Diamond and Dybvig (1983), suppose initially that the return R and the fraction t of early depositors are deterministic, and that the liquidation value of the long-term investment is =1.Then,thebankoffers a deposit contract to depositors so as to maximize U =maxtu 1 (c 11 )+(1 t)u 2 (c 22 ) (1) {c ij } subject to: tc 11 M (2) (1 t)c 22 R(1 M) (3) U 1 (c 11 ) U 1 (c 22 ) (4) U 2 (c 22 ) U 2 (c 11 ), (5) where expression (1) is depositors expected utility with c ij being the consumption of type j at date i; constraints (2) and (3) represent the resource balance constraints at dates 1 and 2 respectively; and conditions (4) and (5) are the incentive compatibility constraints stating that the deposit contract should be designed so that each type of depositors prefers its own withdrawal profile. If depositors are risk averse with RRA > 1, i.e., u 00 (c)/u 0 (c) > 1, the optimal deposit contract satisfies 1 <c 11 <c 22 <R. This result shows that the deposit contract offers insurance to depositors and is Pareto improving relative to the autarkic situation, where individuals invest directly. The insurance provision, however, makes the bank vulnerable to runs. There is a good equilibrium which 6

7 realizes optimal risk sharing when depositors choose the withdrawal decisions embedded in the deposit contract; but there is also a bad equilibrium in which all depositors withdraw their funds prematurely and the bank collapses. The condition c 11 > 1 implies that depositors find it optimal to withdraw if they simply fear that others will withdraw first. There is no rational motivation behind such a panic run other than a coordination failure due to sunspots. The possibility of a run is intrinsic in the provision of insurance. If c 11 1, no run would occur. An alternative explanation for the occurrence of bank runs is that they are linked to changes in fundamental variables and are therefore information-based (or fundamental based). If the return on the long-term investment is stochastic, the perspective of a negative shock increases the probability that the bank is unable to meet its future commitments. If depositors anticipate this, they withdraw their funds and force the premature closure of the bank. Jacklin and Bhattacharya (1988) formalize this mechanism in a context where the assumptions of an illiquid and risky long-term asset and depositors with RRA<1 leadtoan optimal contract satisfying c 11 < 1 <c 22. Whereas this solution excludes the possibility of irrational runs, it still leaves room for information-based runs. After the contract is signed, some type 2 depositors receive a partial signal s describing the posterior distribution of the success probability of the long-term asset. Thus, they update their priors p and choose to withdraw prematurely whenever be U 2 (c 11 ) > E b U 2 (c 22 ), (18) that is whenever the expected utility from withdrawing prematurely calculated using the posterior beliefs on the success probability p, be [U 2 (c 11 )], exceeds the expected utility from waiting and receiving the consumption profile initially designed for them, E b [U 2 (c 22 )]. This triggers a run. Given the total illiquidity of the long-term asset, the bank does not have enough funds to satisfy the withdrawal demands at date 1, and has to close down. The origin of the run is now the rational response of depositors to the arrival of sufficiently negative 7

8 information on the future solvency of the bank. Therefore, the run is information-based and is efficient, as long as it leads to the liquidation of an impending insolvent bank. Panic and information-based runs can also be related, as shown by Chari and Jagannathan (1988). The analysis focuses on the signal-extraction problem faced by uninformed depositors in their withdrawal decisions in a framework characterized by shocks to asset returns and to the proportions of early depositors and informed depositors. Late type depositors who remain uninformed know that other depositors may be informed on the future return of bank asset and try to infer such information from the size of the withdrawal queue at date 1. However, since the proportion of early depositors is stochastic and unobservable, uninformed depositors may not be able to infer bank s future performance correctly. In particular, they may not be able to distinguish whether a long queue is formed by the informed depositors receiving a negative signal, or simply by a large proportion of early depositors wishing to consume early. A pure panic run generates from uninformed depositors confusion between insolvency and high liquidity shocks. It occurs when uninformed depositors withdraw prematurely for fear that some depositors have received a bad signal on the bank s future performance in cases where no one is informed about it. A common feature in this strand of literature is the presence of multiple equilibria, in one of which a bank run occurs. A potential problem with this approach is that individuals may not want to deposit in the first place since they cannot calculate the probability of a run occurring. Consequently, runs should not be observed in equilibrium as no one would deposit anticipating a run. This leaves open the important question of whether the emergence of banks as liquidity providers is desirable from an ex-ante perspective. One way around the multiplicity of equilibria, as suggested by Postlewaite and Vives (1987), is to associate a bank with a sort of Prisoner s Dilemma-type situation in which agents withdraw their deposits for self-interest reasons rather than for consumption reasons. In this context, agents do not condition their behavior on any exogenous event, and there is only a unique equilibrium involving a positive probability of a bank run. A bank run may occur because depositors have incomplete information about the liquidity shocks they face. 8

9 Another way around the multiplicity of equilibria is suggested by Rochet and Vives (2004) and Goldstein and Pauzner (2005). They analyze a modification of the Diamond and Dybvig model, in which the fundamentals of the economy uniquely determine whether a bank run occurs. The key features of the analysis are the assumptions that fundamentals are stochastic and investors obtain noisy private signals on the realization of the fundamentals. This leads to a unique equilibrium in which a bank run occurs when the fundamentals are below some critical level. Importantly, despite being determined by fundamentals, runs can be also driven by bad expectations. Depositors tend to withdraw prematurely for fear that others will do so. Thus a run may occur even when the economy environment is sufficiently good that a run would not occur if depositors had not had bad expectations on other depositors actions. In this respect the model reconciles the view of bank runs as panics due to coordination failure and the one of runs as being linked to fundamentals. The uniqueness of the equilibrium allows also the determination of the ex ante probability of a bank run. This is increasing in the short-term payment the bank offers and therefore in the risk-sharing embodied in the banking contract. When the short-term asset is set at the autarkic level, only efficient runs occur. Depositors withdraw prematurely only if the long-term return of the bank s asset is lower than its liquidation value. In contrast, when the short-term asset is above the autarkic level, inefficient runs occur when a bank is forced to liquidate the long-term asset even though it has a high expected return. Given this inefficiency, having banks offering shortterm payments above the autarkic level is viable and desirable provided that maintaining the long-term investments till maturity is generally efficient. To sum up, bank runs result from either(both) a coordination failure among depositors or(and) an expectation of poor performance of the bank. Runs may be costly, because they force the interruption of a production process and the premature liquidation of assets. Moreover, runs may trigger a systemic crisis, if they propagate through the economy. A systemic crisis has a narrow and a broad interpretation (De Bandt and Hartmann, 2002). A crisis in the narrow sense refers to a situation in which the failure of one bank, or even only the release of bad news about its state of solvency, leads in a sequential fashion to the failure of numerous other banks or of the system as a whole. A crisis in the broad sense 9

10 includes also the simultaneous failure of many banks or of the whole system as result of a generalized adverse shock. The sequential spread out of failures in a narrow crisis implies a strong spillover effect, defined as contagion, which can take place through contagious runs or domino effects (Schoenmaker, 1996). The former refers to the propagation of a run from a single bank to other banks. As for individual runs, such propagation can be due to sunspots or be information-based. The domino effect refers to the mechanism through which difficulties faced by a single bank spread to others through the payment system and/or the interbank market. If relationships among banks are neither collateralized nor insured against, the distress of one bank may trigger a chain of subsequent failures. Other banks may incur a liquidity or an insolvency problem depending on the intensity of the linkages with the distressed bank, and on the correlation of shocks in the system. The channels of contagious runs and domino effects can work in conjunction as well as independently. In most cases, however, a systemic crisis is the result of the propagation of an individual failure through both of them. Most of the interrelations among banks occur through the payment system. Their internal arrangement determines how individual shocks propagate, and thus the severity of the contagion risk. Depending on the timing and the methodology of settlement, payment systems can be classified in net settlement systems (only net balances are settled and at a certain point in time), pure gross systems (payments between members are settled without netting and a certain point in time), real-time gross systems (payments between members are settled without netting and immediately after every transaction) and correspondent banking (payments are settled bilaterally between a correspondent bank and members of a group of small or foreign banks). Net systems economize on liquidity, but expose banks to contagion because they involve the transfer of asset claims from one location to another. By contrast, gross systems entail high liquidity costs, but do not face any risk of contagion (Freixas and Parigi, 1998). Surprisingly, the academic literature has devoted attention to the issue of contagion and systemic risk only very recently. The former models of individual runs can be read in terms 10

11 of generalized systemic crises, but they are not suited for the analysis of the propagation mechanism of individual failures. The analysis of such mechanism requires models with multiple banks. Rochet and Tirole (1996) examine the domino effect in a model of interbank lending with heterogenous banks. Some banks are good at collecting deposits but have poor investment opportunities, while others have plenty of investment opportunities but needs funds. This leaves room for interbank lending, although it exposes banks to the risk of contagion. If a borrowing bank is hit by a liquidity shock, the lending bank may be negatively affected and be forced to shut down. The survival of the lending bank depends on the severity of the shock affecting the borrowing bank and on the revenues (or losses) of the interbank loan. Clearly, the greater the liquidity shock faced by the borrowing bank, the more likely is the closure of the lending bank. The occurrence of contagious runs and domino effects is analyzed by Allen and Gale (2000a) in an economy where banks hold inter-regional deposits on other banks to insure against liquidity preference shocks. The economy works well and achieves optimal risk sharing when there is no aggregate uncertainty; but it may lead to a systemic crisis when there is an excess aggregate demand for liquidity. In such a case, each bank starts to withdraw deposits from banks in other regions in an attempt to satisfy depositors withdrawal demands and avoid liquidating the long-term assets. This mutual liquidation denies liquidity to the troubled bank, which then experiences a run. Depending on the structure of the interbank market, the individual run propagates to other banks and leads to a systemic crisis. If regions are well connected (complete interbank market), contagion is avoided. If connections among regions are limited (incomplete interbank market) and liquidity shocks are strong enough, contagion arises. In a similar spirit, Freixas et al. (2000) analyze the risk of contagious runs through the payment system when banks are located in different regions and face both liquidity and solvency shocks. The former originate from depositors geographical consumption preferences, while the latter from shocks to the return of bank assets. Depositors have two ways 11

12 to satisfy their wish of consuming in a different location from where they have deposited initially. They can withdraw their funds and transfer cash to the other region, or they can transfer deposits from one bank to another through the payment system. When banks are subject only to liquidity shocks, the economy shows multiple equilibria. Either depositors do not run and the payment system is efficient in reducing the opportunity costs of holding liquid assets; or depositors run and banks have to liquidate the long-term assets (speculative gridlock equilibrium). This latter equilibrium resembles the sunspot equilibrium in Diamond and Dybvig. When banks face also (idiosyncratic) solvency shocks, the stability of the system depends on the architecture of the payment system. As in Allen and Gale (2000a), the closure of an insolvent institution is less likely to generate contagious runs when payment systems are well diversified. Excessive risk taking A second source of instability of the banking system relates to risk-taking on the asset side. As it is well known from agency theory, in a principal-agency relationship the objectives of the involved parties are not perfectly aligned so that the agent does not always act in the best interest of the principal. The problem can be limited by designing appropriate incentive schemes for the agent or by controlling his decisions through costly monitoring. In general though, the divergence of interests will not be completely resolved, at least not at zero cost. Applying these arguments to corporate finance, it is easy to see that there is a misalignment in the objectives of debtholders and firm managers. Even if all parties are utility maximizers, their attitude toward risks diverges. Whereas debtholders bear the downside risk, the manager pursuing shareholders interests benefit from upside potential. Thus, the manager has strong incentives to engage in activities that have very high payoffs but very low success probabilities (Jensen and Meckling, 1976). While this agency problem is present in all leveraged firms, two features of the banking system makes it more severe among banks. First, the opacity and the long maturity of banks assets make it easier to cover any misallocation of resources, at least in the short run. Second, the wide dispersion of bank debt among small, uninformed (and often fully 12

13 insured) investors prevents any effective discipline on banks. Thus, because banks can behave less prudently without being easily detected or paying additional funding costs, they have stronger incentives to take risk than firms in other industries. To illustrate the agency problem between banks and depositors, we use a simple model adapted from Holmstrom and Tirole (1997), Cerasi and Daltung (2000) and Carletti (2004). Consider a two-date economy (T =0, 1), in which at date 0 a bank invests in a project, which yields a return R if successful and 0 if unsuccessful. The success probability of the project depends on the monitoring effort m [0, 1] that the bank exerts. It is p H if the bank monitors, and p L if it does not, with p H >p L, p = p H p L and p H R>1 >p L R. Monitoring is costly; an effort m entails a private cost C(m) = c 2 m2. The choice of the monitoring effort depends crucially on the financing structure of the bank. If it is self-financed, it chooses m so to maximize its expected profit Π = mp H R +(1 m)p L R y c 2 m2, where y represents the return on an alternative safe investment. In this case the first order condition gives m = pr. c By contrast, if the bank raises external funds in the form of debt with promised (gross) return r D, it chooses m so as to maximize The first order condition is then given by Π = mp H (R r D )+(1 m)p L (R r D ) c 2 m2. m = p(r r D). c Clearly, raising deposits reduces the equilibrium monitoring effort. The reason is that the bank has now to share the benefit of greater monitoring with depositors. If the deposit rate is set before m is chosen, increasing monitoring simply raises the probability of repaying depositors without reducing the funding costs. This worsens the bank s incentive and leads to a lower equilibrium effort. 13

14 The need of regulation The vulnerability of banks to runs and systemic crises and the consequent concern for consumers wealth are the main factors justifying the need of regulation and safety net arrangements in the form of deposit insurance and lender of last resort. For example, as shown by Diamond and Dybvig (1983), deposit insurance prevents the occurrence of panic (sunspot) runs without reducing banks ability to transform short-term liabilities into long-term assets. A demand deposit contract with government deposit insurance achieves optimal risk sharing among depositors as unique Nash equilibrium. Government s ability to levy nondistortionary taxes and deposit insurance guarantees induce depositors not to withdraw prematurely. Consequently, bank liquidation policy is independent of the volume of withdrawals, no strategic issues of confidence arise and no bank runs take place. The underlying idea behind the introduction of regulation and safety net arrangements is that runs and systemic crises are inefficient and therefore have to be prevented. Whereas this is always true for panic runs, it may not be the case for information-based runs. These are efficient whenever the liquidation value of the long-term asset is higher than its longterm expected return. Given this distinction, it is important to understand why bank runs occur and eventually how to deal with them. Allen and Gale (1998) analyze the potential costs of bank runs and the need for central bank intervention. In their model, bank runs are information-based events which play the important role of sharing risk among depositors. Their welfare properties depend on the potential costs of early withdrawals. When withdrawing early involves no costs, runs are efficient since they occur only when banks long-term asset returns are low. The optimal deposit contract reaches the first-best solution in terms of both risk sharing and portfolio choice, and regulation is not needed. In contrast, when there are real costs associated with early withdrawals (e.g., because the return of the safe asset is higher within the banking system than outside), bank runs reduce the consumption available to depositors. Then laissez-faire does not achieve the first-best allocation any longer and there is scope for central bank intervention in the form of money injection. If the central bank grants an interest-free loan to banks when runs occur, banks 14

15 can avoid liquidating the safe assets prematurely and depositors receive higher consumption levels. The first-best allocation can then be achieved again by a combination of standard deposit contracts, runs and policy intervention. In a similar spirit, central bank intervention is needed when the long-term asset can be liquidated and traded on market. Bank runs are again costly, and the premature liquidation of long-term assets forces down the price in the market and makes crises worse. The intervention of the central bank is needed to prevent the collapse of asset prices. The issue of the optimal form of central bank intervention have long been debated in the academic literature. According to the classical view of Bagehot (1873), central banks should lend freely at a penalty rate and against good collateral. This should prevent banks from using central bank lending to fund current operations and should guarantee that emergency liquidity loans are extended only to illiquid but solvent banks. This view has been criticized in various ways. First, according to Goodhart (1987), it is virtually impossible, even for central banks, to distinguish illiquidity from insolvency at the time the lender of last resort (LOLR) should act. Banks demanding such assistance are under a suspicion of insolvency since they could otherwise raise funds from the market. Second, it has been argued for example by Goodfriend and King (1988) that there is no need for central bank s loans to individual banks since open market operations are sufficient to deal with systemic liquidity crises. In other words, LOLR should intervene at the macroeconomic level but not at the microeconomic level. This debate is also relevant with respect to the possible consequences that the safety net arrangements can create. If on the one hand both deposit insurance and LOLR may suffice to prevent runs and systemic crises, on the other hand they have side effects and bring in new inefficiencies. For example, they worsen the problem of excessive risk taking and call for further regulatory measures. Both deposit insurance and a systematic use of the lender of last resort induce banks to undertake greater risks since depositors do not have incentives to monitor their banks asset values and can rely on future bailout in case of distress (e.g., Merton, 1977; Boot and Greenbaum, 1993). 15

16 Some of these issues have been recently addressed in formal theoretical models. Rochet and Vives (2004) provide a possible theoretical foundation of Bagehot s view using the global game approach. Their analysis builds on a model of banks liquidity crises with a unique Bayesian equilibrium. At this equilibrium there is an intermediate range of values of the bank s asset in which, due to a coordination failure, depositors may run despite the bank being solvent. Thus, as argued by Bagehot, a solvent bank may face a liquidity problem and be in need of assistance. The likelihood of this happening decreases with the ex ante strength of the fundamentals. The optimal policy consists of prudential measures and ex post emergency loans. Liquidity and solvency regulation can solve depositors coordination problem and avoid the failure of solvent banks, but may be too costly in terms of foregone returns. Thus they need to be complemented by emergency discount-window lending. The optimal policy is richer when bank managers can exert an effort and influence the risk of asset returns since it has to account for the effect it has on bank managers incentives. Depending on the value of the fundamentals, the optimal policy may comprise early closure of solvent banks to prevent moral hazard and emergency liquidity assistance. The relationship between banks moral hazard and optimal central bank intervention is also analyzed by Freixas et al. (2004) in a model in which banks are subject to both liquidity and solvency shocks and can operate under moral hazard either in screening loan applicants or in monitoring borrowers. Given the difficulty to discern banks solvency state, insolvent banks may be able to borrow from the interbank market or from the central bank and gamble for resurrection (i.e., invest in projects with negative net present value). The optimal policy depends on the nature of banks incentive problem. If banks face moral hazard in monitoring borrowers, there is no need for central bank intervention. A secured interbank market suffices to implement the first-best allocation. In contrast, if banks face moral hazard mainly in screening loan applicants, the central bank should provide emergency liquidity assistance but at a penalty rate to discourage insolvent banks from borrowing. Gale and Vives (2002) builds on the time-inconsistency embodied in central bank bailout policy to characterize the optimality of dollarization as a way for devaluing depositors claims 16

17 and avoiding bank failure. The idea is that competitive banking systems lead to excessive liquidation when banks face moral hazard problems. By using dollarization as a credible commitment not to bail out banks, the central bank can then implement the incentiveefficient solution and avoid failures. To sum up, the debate around the optimal central bank intervention centres on the trade-off between the benefits (prevention of crises) and the costs (distortion of incentives and moral hazard problem) of bailing out distressed banks. This trade-off may call for other regulatory measures, such as capital regulation, rate regulation and entry restrictions. These too, though, have been heavily criticized as not being effective or inducing other negative distortions, such as a reduction in competition. The sides effects of regulation are therefore crucial for understanding the role and the importance of competition in the banking sector. 3 Competition in banking Analyzing how competition works in the banking sector and whether it is beneficial is a difficult task. On the one hand, the general argument in favor of competition in terms of cost minimization and allocative efficiency apply to the banking industry. On the other hand, however, the presence of various market failures distorts the functioning of competition and makes the standard competitive paradigms not appropriate for the banking sector. The presence of asymmetric information in corporate relationships and of switching costs and networks in retail banking alters the market mechanism. This creates significant entry barriers, which affect the industry structure and lead to an ambiguous relation between the number of banks and the competitive outcome. We analyze these effects more in detail below. It is worth noting though that also other aspects of the role and specificity of banks affect the working of competition in this sector. For example, the simple fact that banks compete on both sides of the balance sheet may lead to departures from the competitive outcome. Stahl (1988) and Yannelle (1988, 1997) show that when banks compete for both loans and deposits, they may want to corner one market in an attempt to achieve a monopoly on the 17

18 other. Furthermore, the role of banks as financiers of industrial loans may create endogenous entry barriers in both the banking sector and the borrowing industries, thus leading to a natural monopoly in both sectors (González-Maestre and Granero, 2003). Competition under asymmetric information As already mentioned, banks emerge as intermediaries between depositors and borrowers. Thus, their two main functions are to provide insurance to depositors, and to screen and monitor investment projects. The former creates the risk of instability; and the latter creates important informational asymmetries among banks and potential borrowers and among banks themselves, which may distort the competitive mechanism significantly. Broecker (1990) analyzes how competition in the credit market affects the screening problem banks face in the choice of granting loans. The set up is such that firms applying for credit differ in their ability to repay loans, i.e., in their credit-worthiness, and banks perform independent and imperfect screening tests to discern firms quality and decide whether to grant loans. Conditional on their own test results, banks compete with each other by setting a loan rate. Given, however, that screening tests are imperfect, the competitive market mechanism does not work properly in that it leads to a negative externality among banks. Increasing the loan rate above that of the competitor has two opposite effects on the profit of the deviating bank. On the one hand, it increases its profit throughtheusualpriceeffect. On the other hand, it worsens the quality of firms accepting the loan, thus reducing its profit. A firm will indeed accept the least favorable loan rate only after being rejected by all other banks setting more favorable rates; but this implies that the firm has a low credit-worthiness on average. Because of this winner s curse problem, increasing the number of banks performing screening tests decreases the average credit-worthiness of firms, and increases the probability that a bank does not grant any loan. In the limit, the number of active banks is positive and the equilibrium maintain some degree of oligopolistic competition. Similar conclusions are reached by Riordan (1993). Using the theory of common value auctions, he shows that a higher number of competing banks worsens the informativeness 18

19 of the signal that banks receive on firms loan quality and make them more conservative in granting loans. Both of these two effects are detrimental for social welfare, since they reduce the quality of banks portfolios and lead to the financing of less efficient investment projects. The relationship between the degree of market competition (or integration) and banks screening incentives is also analyzed by Gehrig (1998). In a context where banks use imperfect credit-worthiness tests to discriminate between good and bad projects, he shows that screening incentives increase with the profitability of loans. Thus, more intense competition due to the entry of outside banks worsens the quality of banks portfolios, since it reduces the investment that banks make to improve the precision of their screening tests. Besides acquiring information on borrowers through screening, banks monitor them also in the course of the relationship, thus obtaining further information on their quality. This creates an informational asymmetry among banks. If a borrower needs a renewal of the loan, the incumbent bank has better information about his quality relative to outside banks. This gives the incumbent bank an informational monopoly over its borrowers, which reduces competition from outside banks, and allows the incumbent bank to hold-up its borrowers and extract monopoly rents. Such expropriation disincentives the borrower from exerting more effort, thus reducing the expected return of investment projects (Rajan, 1992) and leading to an inefficient allocation of capital toward lower quality firms (Sharpe, 1990). The heterogeneity of borrowers and the consequent informational advantages of incumbent banks affect the competitive market mechanism in several ways. As already mentioned, an increase in the number of competing banks reduces the screening ability of each of them. Consequently, more low-quality borrowers obtain financing, and banks may have to increase loan rates to compensate for the higher portfolio risk, thus leading to an inverse relationship between competition and level of loan rates (Marquez, 2002). This results may not obtain any longer, however, when information acquisition is endogenous. In such a context, competition lowers loan rates, in the usual way. Hauswald and Marquez (2005) show that when banks acquire information to soften competition and increase market shares, a higher number of banks reduces the winner s curse problem originating from competitors superior 19

20 information, thus leading to lower loan rates. In other words, an increase in the number of competing banks reduces the degree of product differentiation among banks, and thus loan rates. The presence of adverse selection affects also the structure of the industry. The informational advantage of the incumbent banks allows them to reject the riskier borrowers in need of refinancing. Because outside banks cannot distinguish between new borrowers and old riskier borrowers rejected by their previous incumbent banks, they face an adverse selection problem which may keep them out of the market. An equilibrium of blockaded entry may then emerge, where only two banks are active and make positive profits even under pure Bertrand price competition (Dell Ariccia et al., 1999); or, more generally, the equilibrium is characterized by a finite number of banks even in the absence of exogenous fixed costs (Dell Ariccia, 2001). The general idea is that the heterogeneity of borrowers and the acquisition of information gathered through lending generate endogenous fixed costs, which limit the number of active competitors. To sum up, focusing mostly on an adverse selection problem (i.e., heterogeneity of borrowers), the literature on competition with asymmetric information discusses the possibility for lenders to exercise market power, the imperfect functioning of competitive markets, and the endogenous entry barriers which the informational advantage of incumbent banks can generate. Despite not addressing directly the consequences for stability, this literature provides some intuitions for the effectthatcompetitionmay haveonbanks solvency. Because banks screening abilities worsen with the number of competing banks, tougher competition leads to riskier banks portfolios and high failure probabilities. The mechanism behind the negative relationship between competition and stability derives exclusively from the heterogeneity of borrowers. This contrasts sharply with the mechanism in the literature on competition and stability, where the focus in on how competition modifies the behavior of either borrowers or banks. We will come back to this issue in Section 4. Competition and switching costs 20

21 Switching costs are an important source of market power in retail banking (e.g., Diamond, 1971). In moving from one bank to another, consumers may incur costs associated with the physical change of accounts, bill payments or lack of information (Vives, 2001). The presence of switching costs produces in general two opposing effects on the degree of competition. On the one hand, they may lead to collusive behavior once banks have established a customer base which remains locked in. On the other hand, they induce fierce competition to enlarge the customer base. Thus, switching costs may lead banks to offer high rates initially to attract customers and to reduce them subsequently, when consumers are locked in. Adifferent result may be obtained when switching costs are combined with asymmetric information about borrowers credit-worthiness. Bouckaert and Degryse (2004) analyze a two-period model where heterogenous borrowers face switching costs of changing banks, and banks face an adverse selection problem. In such a context, banks find it convenient to disclose their private information about borrowers credit-worthiness and induce them to switch bank in order to soften overall competition. Disclosure of borrowers quality removes the information disadvantage of rival banks in the interim period, thus allowing them to poach only good borrowers and have positive second-period profits. This relaxes the initial competition for enlarging the customer base, and it increases banks overall profits. Thus, the removal of future informational entry barriers may emerge for strategic reasons as it softens overall competition. The presence of switching costs can also affect significantly the link between number of banks and degree of market competition. Allen and Gale (2000b, 2004) show that a small fixed cost of switching banks may imply higher rates in a system with many small independent banks (unitary system) than in a system with two large banks having extensive nationwide branching networks (branching system). This result is obtained in a model characterized by fixed costs of switching banks, customers initial limited information about the future offer of banks services and prices, and product diversity in the services that banks provide at different locations. Consumers are allocated randomly at each location every period and have to choose which bank to patronize. In a unitary system, each bank consists of one 21

22 branch in one location. Thus, each bank can raise its initial rate by a small amount without losing its customers, because of the fixed cost of switching. The only equilibrium is when all banks charge the monopoly rate. In a branching system, there are only two banks with onebranchineachofthelocations. Althoughconsumerschangelocationineachperiod, they can stay with the same bank if they wish. This possibility increases the costs for each bank of deviating from the equilibrium strategy and losing customers. As a result, branching banking supports more efficient equilibria, where the two banks may charge a rate close to the perfectly competitive level. Competition and networks A final important element affecting the nature of competition in retail banking is the presence of networks. This introduces elements of non-price competition in the interaction between banks, thus affecting the pricing of banking products and the structure of the industry. The possibility for banks of sharing Automatic Teller Machine (ATM) networks can be used as strategic variable to affect price competition. Matutes and Padilla (1994) analyze this issue in a two-period model, where banks choose first whether to build compatible ATM networks and then compete imperfectly on the deposit market. A large ATM network has two opposite effects. On the one hand, it allows banks to offer lower deposit rates, because depositors benefit from an easier access to their deposits when they need cash unexpectedly (network effect). On the other hand, a large ATM network increases price rivalry, because it makes banks more substitutable. Depositors benefit from the location of a bank ATM and the high rates offered by a rival bank sharing the same network (substitution effect). Banks choose to share ATM networks when the network effect dominates, i.e., when depositors do only a small number of transactions through ATMs. The equilibrium is characterized by either partial sharing of ATM networks or no sharing. The former emerges when the network effect dominates; the latter when the substitution effect prevails. Full sharing does not occur in equilibrium, since banks prefer to maintain some differentiation and face softer competition in the deposit market. However, if future entry is possible, banks can use sharing 22

23 agreements to exclude rivals from the market when the network effect is sufficiently high. Then, the threat of entry may lead all incumbents banks to share their network, since this allows them to credibly commit to fierce post-entry competition and foreclose any potential entrant. Sharing of networks is used to limit competition also in McAndrews and Rob (1996) in a two-period model where banks choose whether to own jointly the switches in ATM networks and then compete on the pricing of ATM services. Given the presence of fixed costs in operating a switch and network effects in the demand of ATM services, banks prefer to join switches as a way to achieve a more concentrated structure in the switches industry and monopoly prices in the sale of ATM services to consumers. The implications in terms of welfare are ambiguous. Whereas the joint ownership is inefficient because it leads to the extraction of monopoly rents from final consumers, it is beneficial in that it saves the fixed costs of setting up a switch and gives consumers the possibility of benefitting from a larger network. Similar results are obtained in a context where banks have to decide first whether to offer remote access to their customers, such as postal or telephone services, and then compete for deposits (Degryse, 1996). Depositors differ in terms of taste over location and quality (remote access). Thus, as in Matutes and Padilla (1994), the decision of a bank to offer remote access has the double effect of introducing vertical differentiation between banks and reducing the degree of horizontal differentiation. Consumers with a higher taste for remote access have lower transportation costs if this access is available. Thus, introducing remote access produces two opposite effects. It steals depositors from the rival bank (stealing effect), but it also increases the substitutability between banks (substitution effect). The equilibrium depends on which of these two effects prevails. For low and high values of the ratio quality difference to transportation cost, only one bank offers remote access and offers lower deposit rates. The impact of networks on the structure of the industry and the possibility of entry is analyzed by Gehrig (1996) in a model that also applies to the banking sector, despite being 23